Full 3D Helmholtz Stereopsis Datasets

OVERVIEW

This page contains download links to the synthetic and real datasets used in the paper "A Family of Approaches for Full 3D Reconstruction of Objects with Complex Surface Reflectance". Before you download the datasets, you must first accept the licence conditions below.

LICENCE AGREEMENT

The datasets are freely available under the following terms and conditions:

1. All original images and associated data provided may be used for non-commercial research purposes only.
2. The source of the datasets must be acknowledged in all publications where they are used. This should be done by referencing all of the following:
   • The dataset DOI: 10.15126/surreydata.900004
   • This webpage: https://cvssp.org/data/HS_DS
   • The corresponding publication: Gianmarco Addari and Jean-Yves Guillemaut. A Family of Approaches for Full 3D Reconstruction of Objects with Complex Surface Reflectance. International Journal of Computer Vision, 2023
3. The data may not be redistributed.

DOWNLOAD

DATASET DESCRIPTION

The datasets contain two synthetic scenes ("Bunny" and "Armadillo") rendered using POV-Ray and six real scenes ("Bee", "Fox", "Corgi", "Duck", "Llama" and "Giraffe") captured using a pair of Canon EOS 5DS cameras fitted with Canon Macro Ring Lite MR-14 EX II flashes. All scenes are imaged under reciprocal conditions to allow scene reconstruction using Helmholtz stereopsis.

Data for each scene includes the input reciprocal image pairs ("input??.png" files indexed such that 00-01 form the first reciprocal pair, 02-03 the second reciprocal pair, etc), a silhouette or trimap for each view ("silhouette??.png" files with same index as the corresponding input image) and the calibration information for each view ("input??.txt" or "input??.xml" files with same index as the corresponding input image). The real scenes include an extra image for each view ("input??F.png" files), where the object is lit by the flash mounted on the camera from which the scene is acquired (used only for segmentation purposes). For each scene, we also provide a bounding box ("input_bbox.txt" file) that provides the extent of the scene to be reconstructed.

All views are calibrated with the calibration data stored in .txt files in the form of the projection matrix for the synthetic scenes, and as .xml files in the form of OpenCV compliant calibration data for the real scenes.

Ground truth reconstructions are provided for the synthetic scenes ("Bunny" and "Armadillo") and for three of the real scenes ("Fox", "Corgi" and "Llama"), with the latter obtained from laser scanning. Ground truth reconstruction is not available for the other real scenes which are either made of deformable materials ("Duck" and "Giraffe") or contain moving parts ("Bee"). When available, the ground truth reconstruction is provided as a mesh ("GT.off" file). In the case of the real scenes, the provided ground truth meshes have been scaled and manually registered to the set of images. We recommend performing a further ICP alignment against any reconstruction produced before computing performance metrics.

The models used to render the synthetic scenes are the Stanford Bunny and Armadillo from:

• Greg Turk and Marc Levoy, Zippered Polygon Meshes from Range Images, Proc. SIGGRAPH'94, pages 311–318, 1994
• Venkat Krishnamurthy and Marc Levoy, Fitting Smooth Surfaces to Dense Polygon Meshes, Proc. SIGGRAPH'96, pages 313–324, 1996

available at: http://graphics.stanford.edu/data/3Dscanrep/

SYNTHETIC SCENES

REAL SCENES